In current voice conferencing systems, a speaker selection algorithm in a conferencing bridge detects active speakers and creates an output stream by mixing the audio for the active speakers or active participants. The active stream is then communicated to the participants on the conference call. However, selection of the active speakers involves selecting a predetermined number of most active speakers, commonly referred to as the dominant speakers, based on energy levels of voice communications received from the telephony endpoints where the active speakers are located. All other speakers are excluded from the speaker selection algorithm when speech from the dominant speakers is received.
For example, United States Patent Application publication US20071263821 A1 describes a method and apparatus to provide speaker selection in a multi-party conference call. The method comprises processing a speaker queue for at least one new speaker and monitoring when a number of dominant speakers is less than a predetermined number of dominant speakers. When the predetermined number of dominant speakers is less than the predetermined number, the method automatically, without human intervention, adds the new speaker from the speaker queue to the dominant speakers.
Conventional speaker selection algorithms by design end up not allowing new speakers to join until one of the dominant speakers has been quiet for a while. Although this eliminates interruptions it also precludes new speakers from the opportunity to speak if the dominant speakers continue to keep speaking